[AndyLi429]

Motivation

DeepSeek V4 Flash with DSpark fails during NPU draft graph capture in [this CI job](https://github.com/sgl-project/sglang/actions/runs/38233348303/job/114772999125).

MarkovSampleStep calls -libdevice.log1p(-uniform), but the Ascend backend does not map the generic Triton libdevice stub to its CANN implementation. The call returns None, causing compilation to fail with:

AST unary operator '__neg__' is not (currently) implemented on type NoneType

The sampling kernel was introduced in #43621 and integrated into draft graph capture in #43668.

Modifications

• Import triton.language.extra.cann.libdevice on NPU in markov_sample.py.
• Retain the existing generic libdevice import on other platforms.

Accuracy Tests

• Pre-commit checks passed.
• Host validation with mocked platform/Triton imports and real CPU PyTorch passed: library selection, greedy special values, deterministic sampling, and noise-offset advancement.
• Pending follow-up: validate NPU compilation, graph capture/replay, and rerun the failing accuracy job.

Speed Tests and Profiling

Pending follow-up: verify NPU inference performance after restoring successful compilation. No performance improvement is claimed.

Checklist

• Format your code according to the Format code with pre-commit.
• Add unit tests according to the Run and add unit tests.
• Update documentation according to Write documentations.
• Provide accuracy and speed benchmark results according to Test the accuracy and Benchmark the speed.
• Follow the SGLang code style guidance.

Review and Merge Process

1. Ping Merge Oncalls to start the process. See the PR Merge Process.
2. Get approvals from CODEOWNERS and other reviewers.
3. Trigger CI tests with comments or contact authorized users to do so.
   • Common commands include /tag-and-rerun-ci, /tag-run-ci-label, /rerun-failed-ci
4. After green CI and required approvals, ask Merge Oncalls or people with Write permission to merge the PR.

---

CI States

Latest PR Test (Base): ✅ Run #38245525089
Latest PR Test (Extra): ❌ Run #38245524642
Latest PR Test (AMD ROCm 10): ❌ Run #38245525423

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-12T12:26:51.591965Z	8e4fd24	Draft marked ready

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[sglang-npu-bot]

/tag-and-rerun-ci

[AndyLi429]

fix base-c-test-acc-16-npu-a3 test_npu_deepseek_v4_flash_w8a8_8p_gpqa.py

[AndyLi429]

/rerun-failed-ci

[AndyLi429]

/rerun-failed-ci